Intraindividual Variability Is a Fundamental Phenomenon of Aging: Evidence From an 8-Year Longitudinal Study Across Young, Middle, and Older Adulthood

Abstract

Moment-to-moment intraindividual variability (IIV) in cognitive speed is a sensitive behavioral indicator of the integrity of the aging brain and brain damage, but little information is known about how IIV changes from being relatively low in young adulthood to substantially higher in older adulthood. We evaluated possible age group, sex, and task differences in IIV across adulthood using a large, neurologically normal, population-based sample evaluated thrice over 8 years. Multilevel modeling controlling for education, diabetes, hypertension, and anxiety and depressive symptoms showed expected age group differences in baseline IIV across the adult lifespan. Increase in IIV was not found until older adulthood on simple tasks but was apparent even in the 40s on a more complex task. Females were more variable than males but only at baseline. IIV in cognitive speed is a fundamental behavioral characteristic associated with growing older, even among healthy adults.

Full text

1
Appears in Developmental Psychology (Accepted version)
February 11, 2013
Running Head: INTRAINDIVIDUAL VARIABILITY ACROSS ADULTHOOD
Intraindividual variability is a fundamental phenomenon of aging: Evidence from an 8-year
longitudinal study across young, middle, and older adulthood
Allison A. M. Bielak1*, Nicolas Cherbuin2, David Bunce3, & Kaarin J. Anstey2
1 Department of Human Development and Family Studies, Colorado State University, Fort
Collins, USA
2 Centre for Research on Ageing, Health and Wellbeing, The Australian National University,
Canberra, Australia
3Institute of Psychological Sciences, Faculty of Medicine and Health, University of Leeds, Leeds,
UK
Corresponding author – A. Bielak*
Department of Human Development and Family Studies
1570 Campus Delivery
Colorado State University
Fort Collins, Colorado 80523-1570
Ph: (970) 491-7608
Fax: (970) 491-7975
allison.bielak@colostate.edu

2
Abstract
Moment-to-moment intraindividual variability (IIV) in cognitive speed is a sensitive behavioural
indicator of the integrity of the aging brain and brain damage, but little information is known
about how IIV changes from being relatively low in young adulthood to substantially higher in
older adulthood. We evaluated possible age group, sex, and task differences in IIV across
adulthood using a large, neurologically normal, population-based sample evaluated thrice over 8
years. Multilevel modeling controlling for education, diabetes, hypertension, and anxiety and
depressive symptoms showed expected age group differences in baseline IIV across the adult
lifespan. Increase in IIV was not found until older adulthood on simple tasks, but was apparent
even in the 40s on a more complex task. Females were more variable than males, but only at
baseline. IIV in cognitive speed is a fundamental behavioural characteristic associated with
growing older, even among healthy adults.
Key words: Intraindividual variability, inconsistency, adulthood, change, longitudinal

3
The length of time needed to respond to a stimulus, or reaction time (RT), consistently
increases across adulthood (e.g., Fozard, Vercruyssen, Reynolds, Hancock, & Quilter, 1994).
However, individuals not only become slower with older age, they also become more variable in
their responding from one moment to the next, or from one RT trial to another. A number of
cross-sectional studies have demonstrated that intraindividual variability (IIV) in cognitive speed,
or inconsistency, shows a U-shaped curve across the lifespan (6-89 years, Li et al., 2004; 6-81
years, Williams, Hultsch, Strauss, Hunter, & Tannock, 2005; 5-76 years, Williams, Strauss,
Hultsch, & Hunter, 2007), and is greater among older than younger adults (e.g., Hultsch,
MacDonald, & Dixon, 2002) even when controlling for group differences in mean response speed
and practice effects.
Further, inconsistency in cognitive speed appears to be maladaptive in older age. Greater
variability has been associated with poorer levels of performance on a range of cognitive tasks
and intelligence (e.g., Rabbitt, Osman, Moore, & Stollery, 2001), poorer physical performance
(Anstey, 1999; Li, Aggen, Nesselroade, & Baltes, 2001), less activity participation (Bielak,
Hughes, Small, & Dixon, 2007), poorer performance on tests of everyday functioning (Burton,
Strauss, Hultsch, & Hunter, 2009), and poorer mental health (Bunce, Handley, & Gaines, 2008).
Patients with various types of neurological trauma or disease have been found to show greater IIV
than healthy older adults including those with Parkinson’s disease (Burton, Strauss, Hultsch,
Moll, & Hunter, 2006; de Frias, Dixon, Fisher, & Camicioli, 2007), dementia (Hultsch,
MacDonald, Hunter, Levy-Bencheton, & Strauss, 2000; Murtha, Cismaru, Waechter, &
Chertkow, 2002), mild cognitive impairment (MCI) (Christensen et al., 2005; Dixon et al., 2007;
Strauss, Bielak, Bunce, Hunter, & Hultsch, 2007), and traumatic brain injury (Stuss, Murphy,
Binns, & Alexander, 2003).

4
Greater inconsistency in adulthood has also been linked to maladaptive structural,
functional, and neuromodulatory brain characteristics (see MacDonald, Li, & Bäckman, 2009 for
a review). For example, lower white matter integrity, including decreased volume and increased
prevalence of hyperintensities, have been associated with increased IIV across adulthood
(Anstey et al., 2007; Bunce et al., 2010; Bunce et al., 2007; Fjell, Westlye, Amlien, & Walhovd,
2011; Walhovd & Fjell, 2007). Studies using computational models (Li, Lindenberger, &
Sikström, 2001) and positron emission tomography (MacDonald, Karlsson, Rieckmann, &
Nyberg, 2012) also indicate that dysfunctional dopamine modulation is particularly linked to
more behavioural IIV across adulthood. Overall, although the exact determinants of IIV are not
clearly understood, there is considerable evidence demonstrating the neurological basis of IIV in
adulthood.
Longitudinal studies have shown IIV covaries with cognitive performance across time
(Bielak, Hultsch, Strauss, MacDonald, & Hunter, 2010b; Lövdén, Li, Shing, & Lindenberger,
2007; MacDonald, Hultsch, & Dixon, 2003), and baseline inconsistency predicts later attrition,
mild cognitive impairment (Bielak, Hultsch, Strauss, MacDonald, & Hunter, 2010a; Cherbuin,
Sachdev, & Anstey, 2010), and even death (MacDonald, Hultsch, & Dixon, 2008). Therefore,
inconsistency in adulthood is believed to be a behavioural indicator of neurological integrity,
where compromised integrity translates into less consistent responding on measures of cognitive
speed (Hultsch, et al., 2000; Hultsch, Strauss, Hunter, & MacDonald, 2008).
IIV across the lifespan
Due to the possibility that IIV may be a marker of the integrity of the brain, the majority
of studies have focused exclusively on older adulthood. There has therefore been little
information about how inconsistency changes from being relatively low in young adulthood to
substantially higher in older adulthood. Our aim in the present study was to provide a thorough

5
examination of how inconsistency changes across adulthood using a large, population-based
sample evaluated over 8 years. Previous cross-sectional data covering the adult life-span has
shown that IIV on a choice RT task appears to be at its lowest point in the late teens and 20s,
before steadily increasing with age (Williams, et al., 2005; Williams, et al., 2007). A large cross-
sectional study of adults aged 18-94 years found IIV on a simple RT task remained stable until
approximately age 50, while IIV for choice RT increased linearly across age (Der & Deary,
2006). A similar trend was evident in the first wave of the present study under investigation,
where only those between 60-64 years showed greater inconsistency than those between 40-44
years on a simple RT task, but a stepwise increase was found beginning in young adulthood for
choice RT (Anstey, Dear, Christensen, & Jorm, 2005).
However, cross-sectional age differences do not always correspond to actual changes with
age (e.g., cognitive decline, Salthouse, 2009; Sliwinski & Buschke, 1999). Does IIV increase
over time even among those in their 20s, or is it stable until a certain age? Deary and Der (2005)
tested approximately 500 adults aged 16, 36 and 56 years twice across a 8 year period on simple
and choice RT tasks. Although they initially found inconsistency on both tasks increased linearly
with age, after controlling for mean RT the age effects were dramatically reduced and differences
between the cohorts disappeared. Fozard and colleagues (1994) examined a similar number of
adults aged 20 to 90 years and noted a significant age-related increase in variability in responding
to an auditory choice RT task over 4 years, but did not further describe the nature and shape of
the increase. In an investigation over 6 years, MacDonald and colleagues (2003) found only
those between 75 and 89 years showed significant increases in IIV, while those between 55 and
64 years and between 65 and 74 years remained stable or decreased slightly. Results from other
longitudinal studies focusing exclusively on older adults confirm an increased acceleration in old-
old adulthood (i.e., after 75 years of age, Bielak, et al., 2010b; Lövdén, et al., 2007). Overall,

6
given the conflicting findings in older age, and the limited studies focusing on the earlier half of
adulthood, it is unclear what the longitudinal inconsistency relationship looks like in young and
middle adulthood.
Gender effects
Another unknown factor is whether sex plays a role in age-related IIV change. Females
have been reported to be slower in responding to RT tasks than males (Fozard, et al., 1994; Jorm,
Anstey, Christensen, & Rodgers, 2004), and also show more variability in responding to a choice
RT task across adulthood (Der & Deary, 2006). Others have found this sex difference in IIV
might only be present for those in their 30s and mid adulthood, but not among those in their late
teens and 20s (Deary & Der, 2005). However, Reimers and Maylor (2006) have suggested this
difference might actually be an artifact of failing to account for trial effects. They found females
were slower than males only on the initial trials of responding, but became faster than males
across the RT task. When the initial trials were excluded, the sex difference in inconsistency
disappeared. Therefore, it is unknown whether males and females truly differ in their
inconsistency in cognitive speed.
Cognitive load and IIV
Finally, both age and sex differences in IIV might also vary by the complexity of the
reaction time task. In a 3 year longitudinal study of over 300 participants aged 64 to 92 years,
Bielak and colleagues (2010b) found individuals older than 75 years of age showed significantly
greater annual increases in inconsistency for measures derived from choice RT and task-
switching RT tasks, but not for inconsistency from a simple tapping RT task. Other examples of
similar task-related group differences in IIV abound in the literature, where larger effects have
been found for tasks drawing on executive processes such as inhibition, task switching, or
working memory (e.g., Dixon, et al., 2007; MacDonald, et al., 2003; West, Murphy, Armilio,

7
Craik, & Stuss, 2002). Consequently, the pattern of change in inconsistency across time may
differ based on task complexity, and affect the detection of age and sex differences.
In the present study, we evaluated change in IIV across 8 years in a population-based
sample of 3 different age cohorts, who at baseline were between 20-24 years, 40-44 years, and
60-64 years. This unique longitudinal study provided an excellent design through which to
examine possible age group, sex, and task differences in inconsistency throughout the adult
lifespan. We investigated whether there were significant age group, sex, and age group by sex
differences in the starting value of and change in IIV on two different RT tasks over 8 years.
Because past research on both age and sex differences in IIV conflict, we limited our hypotheses
to the predictions that inconsistency would increase with age, and that age differences would be
greater on a more cognitively challenging RT task. Finally, because higher IIV has been linked
to poorer cognitive performance and various medical conditions (Bunce, et al., 2008; Whitehead,
Dixon, Hultsch, & MacDonald, 2011), we controlled for education, diabetes, hypertension, and
anxiety and depressive symptoms.
Method
Data were drawn from the PATH Through Life Project (PATH), a longitudinal study
whereby participants from 3 different age cohorts (i.e., 20s, 40s, and 60s) are repeatedly tested
every 4 years (see Anstey et al., 2011). The current analyses use three waves of testing (i.e., over
8 years).
Participants
PATH participants are community-dwelling adults residing in the city of Canberra or the
neighbouring town of Queanbeyan, Australia. Potential participants included those aged 20-24
years on January 1, 1999, 40-44 years on January 1, 2000, and 60-64 years on January 1, 2001.
Participants were recruited through the electoral rolls, for which registration is compulsory for

8
Australian citizens. The number of participants who returned the survey totaled 7, 485, of whom
2,404 were in the 20s, 2,530 in the 40s, and 2,551 in the oldest cohort. Approximately half of
each age cohort was female.
There was limited sample attrition 4 and 8 years later, as 6,680 and 5,996 participants
completed Waves 2 and 3, respectively. Participants were excluded from the present analyses if
they reported having a history of stroke, significant head injury, epilepsy, Parkinson’s disease, or
brain tumor, and older participants who scored less than 24 on the Mini-Mental State
Examination (MMSE; Folstein, Folstein, & McHugh, 1975) at any time point. Participants had to
have valid data for all covariate measures and sufficient RT data (see intraindividual variability
section), resulting in a final sample of 6562 participants. The mean length of follow-up among
participants was 6.91 years (SD = 2.54). Further descriptive information about the sample is
presented in Table 1.
Measures and procedure
At each wave, participants answered a questionnaire that assessed their sociodemographic
characteristics, and completed measures of well-being, mental and physical health, and cognitive
functioning. The majority of the assessment was administered on a hand-held or laptop
computer, and was completed under the supervision of and with the assistance of a trained
interviewer (for further details see Anstey, et al., 2011).
Intraindividual variability.
Intraindividual variability was calculated from the response latencies on two reaction-time
tasks, each administered once per testing wave. Both tasks were completed using a small box
which served as both the response console and the display area. The box was held with both
hands, with left and right buttons at the top to be depressed by the index fingers. The front of the
box had three lights: two red stimulus lights under the left and right buttons respectively and a

9
green get-ready light in the middle beneath these. The simple reaction-time (SRT) task was
completed first, immediately followed by the choice reaction-time (CRT) task. In SRT,
participants were presented with a green get-ready light, followed by the right red light after
varying amounts of time. Participants were asked to press a button as soon as the red light
appeared. For each CRT trial, participants were presented with the green get-ready light. After
varying amounts of time, one of the two red lights illuminated and participants were asked to
press the corresponding response button as soon as possible. There were 40 trials presented for
CRT, and 80 for SRT.
Covariates.
We chose to control for the effects of education, diabetes, hypertension, and anxiety and
depressive symptoms. Education was assessed by years of formal schooling (M = 14.86, SD =
2.31), and diabetes was based on the self-reported presence of the disease at any wave (5.7% of
sample). Hypertension was determined from blood pressure readings administered by testers at
each wave, and any participant scoring above 140 systolic or 90 diastolic, or reporting taking
blood pressure medication at any wave was coded as having hypertension (48.2% of sample).
Anxiety and depressive symptoms were based on responses to the Goldberg Anxiety and
Depression Scale (Goldberg, Bridges, Duncan-Jones, & Grayson, 1988), and were entered into
the models separately as time-varying anxiety and depression scores.
Data preparation and calculation of intraindividual variability
The data preparation and intraindividual variability procedures were completed separately
for each task at each wave. Individuals who did not complete more than 50% of the trials1 for
each task were removed from the IIV calculation for that wave. First, incorrect CRT responses
1 A significant reduction in the accuracy of imputation has been shown above 50% of item-level missingness (Burns
et al., 2011).

10
were removed to ensure that IIV was not the result of slower wrong responses. The remaining
RT latencies for both tasks were then trimmed for outliers. Lower limit trims included removing
any trials below 50ms for SRT, and below 150ms for CRT. Means and standard deviations were
next calculated for each individual across all trials, and any latencies that exceeded +3SDs for
that individual were deleted2. Missing values were imputed using a regression substitution
procedure, whereby individual regression equations across all trials for each task were formed
and used to predict the missing values (Hultsch, et al., 2000). This approach reduces within-
subject variation and represents a conservative approach to estimating inconsistency.
Approximately 4% of trials were subjected to the imputation procedure.
The calculation of IIV was in accordance with methodology developed by Hultsch and
colleagues (2000). In order to account for potential confounding influences in the RT data (e.g.,
age differences in mean RT, practice effects), the trial RT data was regressed onto categorical age
group, categorical trial, and their interactions (see Hultsch, et al., 2008 for a description of
statistical considerations). This effectively removed mean RT trends from the data. The resulting
residuals were then converted to standardized T-scores and each individual’s SD across all trials
was calculated. The individual standard deviation (ISD) was used as the indicator of
intraindividual variability. ISD values were computed for each task at each wave (see
Supplementary Table 1). As a further evaluation of this calculation strategy, we found the
additional residualization of individual mean RT trends (i.e., within-person linear trial effects),
followed by the removal of the mean age group and trial effects produced essentially identical
2A previous trimming of the Wave 1 SRT and CRT data for the 60s age group accidentally, but permanently, deleted
the trials that exceeded the trim cutoffs. We employed a different trimming procedure and had to apply our method
to the existing data after that event. Thus, the trimming procedure varied slightly for the Wave 1 60s cohort relative
to the other cohorts and waves.

11
ISD values to our initial calculation approach (r = .999; see supplementary Appendix A for the
statistical equations used in this additional calculation).
Statistical analyses
The ISD data were analyzed using multilevel models which allow the estimation of
individual differences rather than only group differences as in multiple regression. These models
also permit the inclusion of cases with incomplete data, and do not require equal spacing between
waves. Because not all participants were tested at precise 4-year intervals, ISD change was
modeled using a time in study metric. ISDs from the two RT tasks were evaluated separately.
Age group, sex, and age group X sex were included as fixed predictors of the intercept. Models
which included sex predicting the slope were not significant and did not significantly add to the
model fit for either task; therefore only age group was a fixed predictor of the slope in the final
model. The covariates education, diabetes, and hypertension were included as time-invariant
predictors of the intercept, and anxiety and depressive symptoms were included as time-varying
predictors. Random effects for the intercept and slope were estimated for both tasks (see
supplementary appendix B for the statistical equations used), but initial models for SRT indicated
a modest random slope variance (Estimate = .007, SE = .002, p<.01). Further models for SRT
failed to converge, and thus the presented results for SRT do not include random slope. The
results from the unconditional and full models with all factors and covariates included are
presented in Table 2.
Results
ISD change - SRT
There were significant age group, sex, and Age group x Sex interaction effects for
baseline ISD on the SRT task. For all three age cohorts, females tended to have higher ISD
intercepts than males (20s, β = .42, SE = .10, p<.001; 40s: β = .33, SE = .10, p<.01; and 60s: β =

12
.72, SE = .10, p<.001). Although the age group differences were also similar across the sexes, it
appeared that the size of these differences varied by sex. In the 60s cohort both males (M = 6.40)
and females (M = 7.12) showed a higher average starting value than their respective 20s cohort
groups (males: M = 5.14; β = -1.26, SE = .12, p<.001; females: M = 5.56; β = -1.56, SE = .12,
p<.001), but this age difference was significantly larger for females (β = .30, SE = .14, p<.05).
Further, the difference between the 40s (M = 6.15) and 60s males, albeit significant (β = -.25, SE
= .11, p<.05), was less pronounced than for the female 40s (M = 6.48) versus 60s comparison (β
= -.64, SE = .11, p<.001; age group x sex comparison: β = -.39, SE = .14, p<.01). Finally, those
in their 40s also showed a higher baseline ISD than those in their 20s (females: β = -.92, SE = .11,
p<.001; males: β = -1.01, SE = .12, p<.001), but this age comparison did not significantly differ
by sex. Initial models showed a significant amount of between-person variance in intercept
(estimate: 3.92, SE = .11, p<.001). When age group was entered into the model, there was a
19.7% reduction in this intercept variance.
Regarding change over time in study, the oldest cohort showed an average increase in
their inconsistency over time (β = .17, SE = .01, p<.001), while the 40s and 20s age groups both
showed slight decreases (20: β = -.06, SE = .01, p<.001; 40: β = -.06, SE = .01, p<.001; see
Figure 1). Although the 60s group was significantly different from the two younger groups
regarding change over time (both p<.001), the 20s and 40s groups did not differ from one
another.
ISD change - CRT
The effects for baseline CRT were similar to those for SRT, but generally lacking sex
differences, and the Age group x Sex interaction. Only the average female in her 20s (M = 5.93)
was more inconsistent at baseline than the average male of the same age (M = 5.68; β = .25, SE =
.07, p<.01). Regarding cohort differences, the pattern was as expected with both sexes in the 60s

13
cohort showing the highest average baseline ISD (females: M = 7.85; males: M = 7.78) compared
to the average person of the same sex in their 40s (females: M = 6.70; β = -1.09, SE = .08,
p<.001; males: M = 6.76; β = -1.07, SE = .08, p<.001), and 20s (females: β = -1.93, SE = .09,
p<.001; males: β = -2.10, SE = .09, p<.001), and females and males in their 40s having a higher
ISD than the 20s group (females: β = -.83, SE = .08, p<.001; males: β = -1.02, SE = .08, p<.001).
Initial models showed a significant amount of between-person variance in intercept (estimate:
2.62, SE = .07, p<.001), and that the entry of age group into the model accounted for 33.6% of
this intercept variance.
Figure 1 shows that the two oldest cohorts both showed average increases in their
inconsistency over time (60s: β = .16, SE = .01, p<.001; 40s: β = .06, SE = .01, p<.001), but the
20s cohort did not significantly change over time. However, all three groups significantly
differed from one another regarding change over time (all p<.001; see Table 2). Initial models
showed there was a significant, yet modest amount of between-person variation in slope (estimate
= .01, SE = .001, p<.001). Progressive model building showed that age group accounted for 85%
of this slope variance.
Discussion
The main finding of the present study is that increases in IIV are a fundamental
behavioural characteristic associated with growing older, even among healthy adults. There is
sufficient evidence to confidently designate IIV in cognitive speed a developmental phenomenon,
where IIV gradually increases across the adulthood lifespan, showing significant change even in
mid adulthood.
Age group effects
We found a stepwise age group difference in baseline IIV on both simple and choice RT,
where those in their 60s were more inconsistent than those in their 40s, and, in turn, those in their

14
40s were more inconsistent than those in their 20s. These results are in line with previous cross-
sectional work (Hultsch, et al., 2002; Li, et al., 2004; Williams, et al., 2005; Williams, et al.,
2007). Although Der and Deary (2006) and Anstey et al. (2005) found the same pattern cross-
sectionally for IIV on choice RT, both found a relatively flat relationship up until age 50 or 60 for
IIV on simple RT. The use of multilevel models in our study allowed the intercepts to vary by
individual, possibly providing further variance and accuracy to our estimates. Overall, it appears
that baseline inconsistency increases across adulthood, regardless of the complexity of the RT
task.
The pattern is slightly different regarding actual change across time. For the simple RT
task, only the oldest cohort showed a positive slope in IIV across the 8 years, with the two
younger groups both slightly decreasing in IIV over time. In contrast, for choice RT the 40s age
group became slightly more inconsistent with age, and the 60s age group showed even larger
increases in variability. However, those in their 20s still did not significantly change over time.
Previous research has shown the increases in IIV on moderately complex RT tasks (i.e., 2- and 4-
choice RT) and highly complex RT tasks (i.e., 4-choice 1-back RT and 2-choice switch RT) to be
greater with each additional year past age 75 (Bielak, et al., 2010b). Together with our present
results, this suggests that on RT tasks that involve some cognitive complexity (i.e., other than a
simple RT task), adults aged 40 and up show significant increases in variability over time, with
the magnitude of the gain also increasing with greater age (i.e., 60s). However, MacDonald et al.
(2003) found 6-year increases in IIVs for only those between 75 and 89 years at baseline, and
slight decreases or stability for those between 55 and 64 years, and between 65 and 74 years, even
for cognitively challenging RT tasks (i.e., lexical and semantic decision). Therefore, further
longitudinal data across the entire range of adulthood is needed.

15
Although the size of the change that occurred over 8 years was relatively small, the
change in CRT IIV was almost entirely accounted for by age group (a random slope parameter
could not be estimated for SRT). The substantial influence of age is precisely what one would
expect to find if increases in IIV are indeed a developmental phenomenon associated with non-
pathological aging. Further, the size of the changes with each additional year would not be
expected to be very large given the size of the differences between the age groups at baseline
(e.g., those in their 40s cannot increase at a rate that would have them reach the baseline level of
the 60s cohort well before their 60th birthday). Therefore, the size of the changes in IIV with age
must be commensurate with the size of the age differences on the task itself amongst healthy
adults. In addition, given demonstrations that IIV change covaries with cognitive change (e.g.,
Lövdén, et al., 2007), any shifts in IIV should correspond with the size of the cognitive change
expected for that age group. This differential change aligns with our findings that IIV changes
were most prominent for the oldest cohort, who experience a greater rate of cognitive change than
younger adults (e.g., Salthouse, 2009). However, it appears that IIV change in simple RT may
not follow the same rules. Rather, purely process-based IIV may operate on a step-wise rather
than constant function in early and middle adulthood (i.e., with jumps eventually showing
individuals performing at levels consistent with their new age group), and not show consistent
developmental increases until older adulthood.
Given that maladaptive IIV is believed to be a sensitive behavioural indicator of
neurological integrity, it is intriguing that increases in CRT inconsistency over 8 years were seen
even among those aged 40-44 years at baseline. On the other hand, the slow decline of various
cognitive abilities across adulthood, particularly processing speed, is well documented
(Salthouse, 2009). In fact, Bunce and colleagues (2010) found an association between IIV and

16
frontal white-matter hyperintensities in a cohort subsample from the same sample, further
demonstrating that neurological integrity may be compromised well before older age.
Sex, and Age group X Sex interaction effects
We also found evidence of sex differences in IIV, but only in relation to the baseline level
and not change over time. For inconsistency on the simple RT task, females had higher intercepts
than males across all three age groups, and age differences were more pronounced among
females. The sex differences were reduced for choice RT, where only females in the youngest
age group were more inconsistent than males of the same age. Our analyses controlled for trial
effects, and thus do not support Reimers and Maylor’s (2006) suggestion that sex differences in
inconsistency are only the result of females’ slower responding on the initial trials. Although the
direction of the sex difference is the same, our findings are in contrast to past work finding the
strongest sex effect for choice RT and no differences for simple RT (Der & Deary, 2006), and
showing only females aged 36 and older had more variability than males, and only on choice RT
(Deary & Der, 2005).
Although the explanation as to why females were more variable at the first wave of testing
is unclear, the fact that this was only found for simple RT and not choice RT is intriguing. The
simple RT task was completed before the choice RT task, and it may be that aspects of the testing
situation influenced females (e.g., test anxiety) more than males during the simple RT task, but
these factors then diminished during the second RT task. Further, because past research has
consistently found larger IIV group differences on RT tasks that pose a greater cognitive
challenge (e.g., West, et al., 2002), the fact that the same was not found in relation to sex suggests
that the sex-related difference is not substantially related to differences in neurological integrity.
For example, it is for groups that are believed to have poorer neurological integrity relative to
their comparison group where task complexity differences have been found, such as those with

17
mild cognitive impairment (Dixon, et al., 2007). Further, because there was no evidence that
males and females show different change in IIV across time, the presence of the intercept
difference is likely not of neurological interest.
Given the interest in comparing IIV to mean RT (Hultsch, et al., 2008) we additionally
evaluated the size of the model-implied change in mean RT for both tasks across the 8 years. The
trials were converted to T-scores before calculating individual mean RT, thus permitting
comparison with the ISD slope parameter estimates. For SRT3, the 60s cohort showed an average
increase in mean RT over time (60s: β = 2.10), but the two younger cohorts showed significant
decreases (20s: β = -1.49; 40s: β = -.43). The results were similar for CRT4, with the 60s cohort
having slower average responding over time (β = .78) and the 20s responding faster (β = -.33),
but the 40s cohort showing no significant change. Therefore, the general pattern of age-related
differences in change is comparable to that observed in IIV. However, the change per year is
greater for mean RT. Thus, although there has been evidence that IIV and mean RT might be
fundamentally distinct phenomena (Burton, et al., 2009; Lövdén, et al., 2007), both appear to
change relatively similarly across adulthood.
Despite the strengths of the present study, including a population-based sample, the large
number of study participants and the longitudinal design, some limitations must be considered.
First, there were only two indicators of IIV, both derived from relatively simple psychomotor
tasks. Analyses on inconsistency computed from higher-order or more cognitive challenging RT
tasks may demonstrate a different pattern of change. However, given past findings regarding task
complexity (Bielak, et al., 2010b), the pattern using such tasks is predicted to be even stronger,
with more pronounced age differences. Next, although the sample represented three distinct age
3 There was sufficient variation in mean RT change to estimate a random slope parameter.
4 There was insufficient variation in mean RT change, and a random slope parameter could not be estimated.

18
cohorts and provided an estimate of change across young, middle, and older adulthood, the spread
in the age groups prevented a continuous estimate across adulthood. Relatedly, we also were
limited to examining relatively early old age, and IIV change has been shown to be even greater
in the later stages of older adulthood (Lövdén, et al., 2007). Further, the greatest change occurred
between wave 1 and wave 2 for all age groups, with slight group-based decreases in IIV from
waves 2 to 3. Although this pattern implies a quadratic function, we were unable to include this
without model saturation. However, proportionally larger increases in IIV with age were still
evident, reiterating the model-implied conclusion that IIV in processing speed change is a
developmental phenomenon that increases in magnitude with age. Our multilevel models also
permitted individual variation in the intercept (and the slope for CRT), and can provide greater
insight than relying on group-based change from descriptive data. The decrease in IIV was likely
the result of practice effects, with the 20s and 40s cohorts showing the greatest benefit. A fourth
wave of data collection will clarify the longitudinal changes by permitting evaluation of IIV
change from 20-36 years, 40-56 years, and 60-76 years of age. Finally, it remains a possibility
that a portion of age-related changes in IIV could be due to age differences in strategic response
behavior (i.e., the diffusion model) (Ratcliff & McKoon, 2008).
The present study aimed to fill a gap in the knowledge base on how moment-to-moment
IIV in cognitive performance changes across adulthood. It appears that increases in IIV are a
fundamental phenomenon associated with growing older, even among healthy adults. The
magnitude of the increase depends on the task, with IIV on simpler tasks not increasing until
older adulthood, and IIV on more complex tasks showing increases as early as middle adulthood.
Given the documented predictive prowess of IIV in forecasting changes in neurological integrity
(e.g., Bielak, et al., 2010a; MacDonald, et al., 2008), a shift towards greater attention to increases
in inconsistency as markers of biological and cognitive aging, even in mid-life, is warranted.

19
Acknowledgements
We thank the study participants, PATH interviewers, Trish Jacomb, Karen Maxwell, Tony Jorm,
Helen Christensen, Bryan Rodgers, Peter Butterworth and Simon Easteal for their contribution to
the research. K. J. Anstey and N. Cherbuin were supported by National Health and Medical
Research Council (NHMRC) Fellowships (No. 1002560 and 471501, respectively). D. Bunce
was supported by a Leverhulme Trust (UK) Research Fellowship. The PATH Through Life
Study was funded by NHMRC Grants (No. 229936 and 179839).

20
References
Anstey, K. J. (1999). Sensorimotor and forced expiratory volume as correlates of speed, accuracy,
and variability in reaction time performance in late adulthood. Aging, Neuropsychology,
and Cognition, 6, 84-95.
Anstey, K. J., Christensen, H., Butterworth, P., Easteal, S., Mackinnon, A., Jacomb, T., . . . Jorm,
A. F. (2011). Cohort profile: The PATH through life project. International Journal of
Epidemiology. doi: doi:10.1093/ije/dyr025
Anstey, K. J., Dear, K., Christensen, H., & Jorm, A. F. (2005). Biomarkers, health, lifestyle, and
demographic variables as correlates of reaction time performance in early, middle, and
late adulthood. The Quarterly Journal of Experimental Psychology, 58A, 5-21.
Anstey, K. J., Mack, H. A., Christensen, H., Li, S-C., Reglade-Meslin, C., Maller, J., . . .
Sachdev, P. (2007). Corpus callosum size, reaction time speed and variability in mild
cognitive disorders and in a normative sample. Neuropsychologia, 45, 1911-1920.
Bielak, A. A. M., Hughes, T. F., Small, B. J., & Dixon, R. A. (2007). It's never too late to engage
in lifestyle activities: Significant concurrent but not change relationships between lifestyle
activities and cognitive speed. Journal of Gerontology: Psychological Sciences, 62B,
P331-P339.
Bielak, A. A. M., Hultsch, D. F., Strauss, E., MacDonald, S. W. S., & Hunter, M. A. (2010a).
Intraindividual variability in reaction time predicts cognitive outcomes 5 years later.
Neuropsychology, 24, 731-741.
Bielak, A. A. M., Hultsch, D. F., Strauss, E., MacDonald, S. W. S., & Hunter, M. A. (2010b).
Intraindividual variability is related to cognitive change in older adults: Evidence for
within-person coupling. Psychology and Aging, 25, 575-586.

21
Bunce, D., Anstey, K. J., Cherbuin, N., Burns, R., Christensen, H., Wen, W., & Sachdev, P.
(2010). Cognitive deficits are associated with frontal and temporal lobe white matter
lesions in middle-aged adults living in the community. PLoS ONE, 5(10), e13567. doi:
10.1371/journal.pone.0013567
Bunce, D., Anstey, K. J., Christensen, H., Dear, K., Wen, W., & Sachdev, P. (2007). White
matter hyperintensities and within-person variability in community-dwelling adults aged
60-64 years. Neuropsychologia, 45, 2009-2015.
Bunce, D., Handley, R., & Gaines, S. O. Jr. (2008). Depression, anxiety, and within-person
variability in adults aged 18 to 85 years. Psychology and Aging, 23, 848-858.
Burns, R. A., Butterworth, P., Kiely, K. M., Bielak, A. A. M., Luszcz, M. A., Mitchell, P., . . .
Anstey, K. J. . (2011). Multiple imputation was an efficient method for harmonizing the Mini-
Mental State Examination with missing item-level data. Journal of Clinical Epidemiology, 64,
787-793.
Burton, C. L., Strauss, E., Hultsch, D. F., & Hunter, M. A. (2009). The relationship between
everyday problem solving and inconsistency in reaction time in older adults. Aging,
Neuropsychology, and Cognition, 16, 607-632.
Burton, C. L., Strauss, E., Hultsch, D. F., Moll, A., & Hunter, M. A. (2006). Intraindividual
variability as a marker of neurological dysfunction: A comparison of Alzheimer’s disease
and Parkinson’s disease. Journal of Clinical and Experimental Neuropsychology, 28, 67-
83.
Cherbuin, N., Sachdev, P., & Anstey, K. J. (2010). Neuropsychological predictors of transition
from healthy cognitive aging to mild cognitive impairment: The PATH Through Life
Study. American Journal of Geriatric Psychiatry, 18(8), 723-733.

22
Christensen, H., Dear, K. B. G., Anstey, K. J., Parslow, R. A., Sachdev, P., & Jorm, A. F. .
(2005). Within-occasion intraindividual variability and preclinical diagnostic status: Is
intraindividual variability an indicator of mild cognitive impairment? . Neuropsychology,
19, 309-317.
de Frias, C. M., Dixon, R. A., Fisher, N., & Camicioli, R. (2007). Intraindividual variability in
neurocognitive speed: A comparison of Parkinson's disease and normal older adults.
Neuropsychologia, 45, 2499-2507.
Deary, I. J., & Der, G. (2005). Reaction time, age, and cognitive ability: Longitudinal findings
from age 16 to 63 years in representative population samples. Aging, Neuropsychology,
and Cognition, 12, 187-215.
Der, G., & Deary, I. J. (2006). Age and sex differences in reaction time in adulthood: Results
from the United Kingdom Health and Lifestyle Survey. Psychology and Aging, 21, 62-73.
Dixon, R. A., Garrett, D. D., Lentz, T. L., MacDonald, S. W. S., Strauss, E., & Hultsch, D. F.
(2007). Neurocognitive markers of cognitive impairment: Exploring the roles of speed
and inconsistency. Neuropsychology, 21, 381-399.
Fjell, A. M., Westlye, L. T., Amlien, I. K., & Walhovd, K. B. (2011). Reduced white matter
integrity is related to cognitive instability. The Journal of Neuroscience, 31, 18060-18072.
Folstein, M., Folstein, S., & McHugh, P. R. (1975). Mini-Mental State: A practical method for
grading the cognitive state of patients for the clinician. Journal of Psychiatric Research,
12, 189-198.
Fozard, J. L., Vercruyssen, M., Reynolds, S. L., Hancock, P. A., & Quilter, R. E. (1994). Age
differences and changes in reaction time: The Baltimore Longitudinal Study of Aging.
Journal of Gerontology: Psychological Sciences, 49, P179-P189.

23
Goldberg, D., Bridges, K., Duncan-Jones, P., & Grayson, D. (1988). Detecting anxiety and
depression in general medical settings. BMJ, 297, 897-899.
Hultsch, D. F., MacDonald, S. W. S., & Dixon, R. A. (2002). Variability in reaction time
performance of younger and older adults. Journal of Gerontology: Psychological
Sciences, 57B, P101-P115.
Hultsch, D. F., MacDonald, S. W. S., Hunter, M. A., Levy-Bencheton, J., & Strauss, E. (2000).
Intraindividual variability in cognitive performance in older adults: Comparison of adults
with mild dementia, adults with arthritis, and healthy adults. Neuropsychology, 14, 588-
598.
Hultsch, D. F., Strauss, E., Hunter, M. A., & MacDonald, S. W. S. . (2008). Intraindividual
variability, cognition, and aging. In F. I. M. Craik & T. A. Salthouse (Eds.), The
handbook of aging and cognition (3rd ed) (pp. 491-556). New York: Psychology Press.
Jorm, A. F., Anstey, K. J., Christensen, H., & Rodgers, B. (2004). Gender differences in
cognitive abilities: The mediating role of health state and health habits. Intelligence, 32, 7-
23.
Li, S-C., Aggen, S. H., Nesselroade, J. R., & Baltes, P. B. (2001). Short-term fluctuations in
elderly people's sensorimotor functioning predict text and spatial memory performance:
The MacArthur Successful Aging Studies. Gerontology, 47, 100-116.
Li, S-C., Lindenberger, U., Hommel, B., Aschersleben, G., Prinz, W., & Baltes, P. B. (2004).
Transformations in the couplings among intellectual abilities and constituent cognitive
processes across the life span. Psychological Science, 15, 155-163.
Li, S-C., Lindenberger, U., & Sikström, S. (2001). Aging cognition: From neuromodulation to
representation. Trends in Cognitive Sciences, 5, 479-486.

24
Lövdén, M., Li, S-C., Shing, Y. L., & Lindenberger, U. (2007). Within-person trial-to-trial
variability precedes and predicts cognitive decline in old and very old age: Longitudinal
data from the Berlin Aging Study. Neuropsychologia, 45, 2827-2838.
MacDonald, S. W. S., Hultsch, D. F., & Dixon, R. A. (2003). Performance variability is related to
change in cognition: Evidence from the Victoria Longitudinal Study. Psychology and
Aging, 18, 510-523.
MacDonald, S. W. S., Hultsch, D. F., & Dixon, R. A. (2008). Predicting impending death:
Inconsistency in speed is a selective and early marker. Psychology and Aging, 23, 595-
607.
MacDonald, S. W. S., Karlsson, S., Rieckmann, A., & Nyberg, L. (2012). Aging-related increases
in behavioral variability: Relations to losses of dopamine D1 receptors. The Journal of
Neuroscience, 32, 8186-8191.
MacDonald, S. W. S., Li, S-C., & Bäckman, L. (2009). Neural underpinnings of within-person
variability in cognitive functioning. Psychology and Aging, 24, 792-808.
Murtha, S., Cismaru, R., Waechter, R., & Chertkow, H. (2002). Increased variability
accompanies frontal lobe damage in dementia. Journal of the International
Neuropsychological Society, 8, 360-372.
Rabbitt, P., Osman, P., Moore, B., & Stollery, B. (2001). There are stable individual differences
in performance variability, both from moment to moment and from day to day. The
Quarterly Journal of Experimental Psychology, 54A, 981-1003.
Ratcliff, R., & McKoon, G. (2008). The diffusion decision model: Theory and data for two-
choice decision tasks. Neural Computation, 20, 873-922.

25
Reimers, S., & Maylor, E. A. (2006). Gender effects on reaction time varaibility and trial-to-trial
performance: Reply to Deary and Der (2005). Aging, Neuropsychology, and Cognition,
13, 479-489.
Salthouse, T. A. (2009). When does age-related cognitive decline begin? Neurobiology of Aging,
30, 507-514.
Sliwinski, M., & Buschke, H. (1999). Cross-sectional and longitudinal relationships among age,
cognition, and processing speed. Psychology and Aging, 14(1), 18-33.
Strauss, E., Bielak, A. A. M., Bunce, D., Hunter, M. A., & Hultsch, D. F. (2007). Within-person
variability in response speed as an indicator of mild cognitive impairment in older adults.
Aging, Neuropsychology, and Cognition, 14, 608-630.
Stuss, D. T., Murphy, K. J., Binns, M. A., & Alexander, M. P. (2003). Staying on the job: The
frontal lobes control individual performance variability. Brain, 126, 2363-2380.
Walhovd, K. B., & Fjell, A. M. (2007). White matter volume predicts reaction time instability.
Neuropsychologia, 45, 2277-2284.
West, R., Murphy, K. J., Armilio, M. L., Craik, F. I. M., & Stuss, D. T. (2002). Lapses of
intention and performance variability reveal age-related increases in fluctuations of
executive control. Brain and Cognition, 49, 402-419.
Whitehead, B. P., Dixon, R. A., Hultsch, D. F., & MacDonald, S. W. S. (2011). Are
neurocognitive speed and inconsistency similarly affected in type 2 diabetes? Journal of
Clinical and Experimental Neuropsychology, 33, 647-657. doi:
10.1080/13803395.2010.547845
Williams, B. R., Hultsch, D. F., Strauss, E. H., Hunter, M. A., & Tannock, R. (2005).
Inconsistency in reaction time across the life span. Neuropsychology, 19, 88-96.

26
Williams, B. R., Strauss, E. H., Hultsch, D. F., & Hunter, M. A. (2007). Reaction time
inconsistency in a spatial stroop task: Age-related differences through childhood and
adulthood. Aging, Neuropsychology, and Cognition, 14, 417-439.

27
Table 1. Descriptive information about the sample covariates.
Age group
20 40 60
Male Female Male Female Male Female
n = 1018 n = 1175 n = 1034 n = 1220 n = 1045 n = 1070
Measure M (SD) M (SD) M (SD) M (SD) M (SD) M (SD)
Years of
education
15.34 (1.73) 15.56 (1.72) 15.19 (2.22) 14.86 (2.29) 14.55 (2.65) 13.63 (2.57)
Anxiety at
baseline
3.13 (2.59) 4.40 (2.68) 3.24 (2.62) 3.67 (2.70) 1.81 (2.12) 2.44 (2.35)
Depressive
symptoms at
baseline
2.55 (2.30) 3.16 (2.42) 2.23 (2.28) 2.51 (2.42) 1.42 (1.71) 1.69 (1.85)
% Diabetic 0.7 1.3 4.6 3.7 14.7 9.9
% Hypertensive 36.1 9.1 55.7 34.6 82.9 76.7
Note. SD = Standard deviation. Hypertension was defined as scoring above 140 systolic or 90 diastolic, or reporting taking blood
pressure medication at any wave.

28
Table 2. Parameter Estimates from Multilevel Models Examining Age Group and Sex
Differences in Intraindividual Standard Deviations (ISD) Across 8 Years.
ISD
SRT CRT
Parameter Estimate SE Estimate SE
Unconditional Model (df = 3)
Fixed effects
Intercept 5.89***
.03 6.82***
0.02
Random effects
Intercept variance 3.91***
.11 2.72***
.07
Residual variance 5.45***
.08 2.53***
.04
-2LL 84132 71126
AIC 84138 71132
Final Model
Fixed effects
Intercept 6.40***
.13 7.78***
.10
Time .17***
.01 .16***
.01
Age Group contrasts
60 vs. 20 -1.26***
.11 -2.10***
.09
60 vs. 40 -.25 * .11 -1.07***
.08
40 vs. 20
a
-1.01***
.11 -1.02***
.08
Sex .72***
.10 .07 .07
Age Group x Sex contrasts
60 vs. 20 x Sex
a
.30* .14 -.17 .10
60 vs. 40 x Sex -.39** .14 -.02 .10
40 vs. 20 x Sex
a
.09 .14 .19 .10
Time x Age Group contrasts
60 vs. 20 -.23***
.01 -.15***
.01
60 vs. 40 -.23***
.01 -.10***
.01
40 vs. 20
a
.00 .01 -.05***
.01

29
Random effects
Intercept variance 2.97***
.10 1.71***
.05
Slope variance - .001 .00
Residual variance 5.38***
.08 2.35***
.04
-2LL 82646 (df = 16) 68349 (df = 17)
AIC 82678 68383
Note. * p < . 05, ** p < . 01, *** p < . 001. SRT = simple reaction time; CRT = choice reaction
time. 60s cohort, male, served as reference group. All estimates are unstandardized. Years of
education, diabetes, hypertension, anxiety, and depressive symptoms were included as covariates
in the final model. aContrast tested in another analysis using same model but different coding for
age group. Due to insufficient variation, random slope was not estimated for SRT in the final
model.

30
Figure 1. Model implied change in ISD across time as a function of age group and task.
Note: ISD = Intraindividual standard deviation; SRT = Simple reaction time; CRT = Choice
reaction time.
4
5
6
7
8
9
10
012345678
ISD
Year
20-SRT
40-SRT
60-SRT
20-CRT
40-CRT
60-CRT

31